CN106375768B - Video steganalysis method based on intra prediction mode calibration - Google Patents
Video steganalysis method based on intra prediction mode calibration Download PDFInfo
- Publication number
- CN106375768B CN106375768B CN201510438280.9A CN201510438280A CN106375768B CN 106375768 B CN106375768 B CN 106375768B CN 201510438280 A CN201510438280 A CN 201510438280A CN 106375768 B CN106375768 B CN 106375768B
- Authority
- CN
- China
- Prior art keywords
- prediction mode
- calibration
- intra prediction
- video
- ipm
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Expired - Fee Related
Links
Landscapes
- Compression Or Coding Systems Of Tv Signals (AREA)
Abstract
The present invention relates to a kind of video steganalysis methods based on intra prediction mode calibration.The step of this method includes: that video to be checked 1) is carried out intra prediction mode calibration, the intra prediction mode calibration decompresses and carries out second-compressed with 4 × 4 pieces for unit block-by-block, and various intra prediction mode IPM are traversed during second-compressed and calculate corresponding absolute transformed error and SATD；2) intra prediction mode alignment features collection is calculated according to the data recorded in calibration process, the intra prediction mode alignment features collection includes two subsets of IPM transition probability feature set and SATD transfer distance feature set；3) the intra prediction mode alignment features collection is supplied to classifier to be learnt and classified, and then carries out steganalysis.The existing video steganographic algorithm based on intra prediction mode can be effectively detected in the present invention.
Description
Technical field
The invention belongs to the steganalysis field in information security, it is related to a kind of base under H.264/AVC video encoding standard
In the video steganalysis method of intra prediction mode calibration.
Background technique
Present information steganography reaches hidden by being embedded into secret information to seem in normal digital media carrier
The purpose of secret communication.Steganography carrier is very extensive, including text, audio, image, video etc..The audio visual and system of steganography carrier
It is all quite similar with non-steganography carrier to count characteristic, it is difficult to be distinguished by conventional method.Video is gathered around relative to text, audio and image
There are bigger data volume, wider application scenarios.This also means that video steganography carrier possesses bigger steganographic capacity and more
Extensive route of transmission, just due to these advantages, video steganography is increasingly becoming the emphasis of research.
Information steganography technology is many kinds of, is widely used, to protect the safe transmission of important information to provide advantageous guarantor
Barrier.But the technology is not only used by multinational security department, military forces and intelligence agency, also can be by between hostile force information
Spy, terrorist and cult member using the propagation and communication for carrying out various information, plot, reaction speech etc.,
The politics and economic security of country are seriously threatened.Therefore, for fight the Steganalysis of steganography in recent years at
As the emphasis studied both at home and abroad.The basic principle of steganalysis is according to the statistical property of digital media file data or certain
Condition code judges secret information whether has been embedded into carrier, or even attempts to extract containing the secret information in secret carrier.Through toning
Research and development are existing, at present early, the research achievement compared with the research of Steganalysis starting all using text, image as the steganography of carrier
It is very extensive.The research in video steganography field is also more both at home and abroad at present, just continuously emerges outstanding research achievement；
But the related research result of video steganalysis is then seldom, causes the unbalance situation of attacking and defending.Therefore, in order to effectively contain video
Network Communicate Security problem caused by the abuse of steganography needs correlative study of the booster injection to video Steganalysis.
There is the video based on intra prediction mode (IPM:intra prediction mode) in video steganography field
Steganographic algorithm, such algorithm are generally basede on H.264/AVC video encoding standard, that is, pass through the intra prediction of macro block in modification I frame
Mode is to be embedded in secret information.Such as: there is scholar to propose to set certain mapping ruler first for 9 kinds in " 4x4 intra prediction "
Prediction mode (Fig. 1) is mapped to binary number 0 or 1, and under this mapping ruler, further according to secret information and associate(d) matrix is encoded,
It modifies to the intra prediction mode of macro block in I frame.In conjunction with the steganography coding techniques such as matrix coder, such steganographic algorithm energy
Enough reach good steganography safety, also possesses higher steganographic capacity.For the steganographic algorithm based on intra prediction mode, have
Scholar is based on Markov chain and devises a kind of steganalysis method, but the supposed premise of this method is excessively ideal, practical application
In performance it is unsatisfactory, can not effectively detect the video steganographic algorithm based on intra prediction mode.
Summary of the invention
To solve the above problems, the present invention proposes a kind of video steganalysis method based on intra prediction mode calibration,
The existing video steganographic algorithm based on intra prediction mode can be effectively detected.
The present invention carries out video intra-frame prediction model calibration first, i.e., decompresses video to be checked for unit block-by-block with 4 × 4 pieces
And second-compressed, and intra prediction is reselected during second-compressed；Record related data is based in a calibration process
It calculates intra prediction mode and calibrates (IPMC:Intra Prediction Mode Calibration) feature；Then by IPMC feature
Classifier such as support vector machines (SVM:Support Vector Machine) is supplied to for learning and classifying, and then is carried out
Steganalysis.
The overall technical architecture of video steganalysis method of the present invention based on intra prediction mode calibration includes
Three intra prediction mode calibration, IPMC feature calculation and the study of SVM support vector machines and classification big steps, main-process stream is as schemed
Shown in 2, implementation steps include:
Step 1: carrying out intra prediction mode calibration with 4 × 4 pieces for unit, be possibly used for the 4 of insertion in screening I frame first
× 4 pieces, then by these 4 × 4 pieces of block-by-block decompressions and second-compressed, 9 kinds of prediction modes are traversed during second-compressed and are counted
Calculate corresponding absolute transformed error and (SATD:Sum of Absolute Transformed Differences).
Step 2: IPMC feature set being extracted according to the related data recorded in step 1, IPMC feature set includes that IPM transfer is general
Two subsets of rate feature set and SATD transfer distance feature set.IPM transition probability feature set features the IPM after calibration and deviates
The probability of original optimal IPM, and SATD transfer distance feature set features the distance that the SATD after calibration deviates original value.
Step 3: being learnt and classified using classifier such as SVM support vector machines.When using SVM support vector machines,
Kernel function uses gaussian kernel function.If input is training sample video, IPMC feature is used to train classifier template；If defeated
Enter for video to be checked, then match corresponding classifier template first, then is used to classify whether to determine video by IPMC feature
For steganography carrier.
Video steganography method of the invention has the beneficial effect that correlative technology field:
1. the present invention is relative to existing similar steganalysis method, detection performance, which has, to be quite obviously improved.Needle at present
Less to the steganalysis method achievement based on intra prediction mould steganography, existing method supposed premise is excessively ideal, actually answers
Performance in is unsatisfactory.Calibration program in the present invention can accurately portray steganography distortion caused by video, extract
Feature is very sensitive to the non-dominance of intra prediction mode.Even if video insertion rate to be checked is very low, detection performance is still stablized.
2. the present invention solves the problems, such as that second-compressed parameter is difficult to be adapted in previous video calibration scheme.According to investigation,
Entire video is usually carried out decompression and second-compressed by the existing video steganalysis method based on calibration as a whole,
Usually there are problems that second-compressed parameter is difficult to be adapted to compression for the first time causes detection performance sharply to decline.And in the present invention
Calibration program in, decompression is to be carried out with 4 × 4 pieces for unit with second-compressed, and all second-compressed parameters can be from original
It is obtained in beginning input video, there is no be difficult to the problem of being adapted to.
3. the technical solution that the present invention is calibrated in blocks possesses good flexible expansion.The present invention selection with
Block is the consistency that unit carries out that macro block divides when calibration well ensures second-compressed, but also second-compressed parameter is suitable
It is addressed with problem.Based on this thinking, different calibration programs can be formulated to detect different steganalysis algorithms, such as
Motion vector calibration program as unit of macro block etc., it is seen that the present invention possesses good flexible expansion.
Detailed description of the invention
Fig. 1 is H.264/AVC 9 kinds of intra prediction mode schematic diagrames in video encoding standard.
Fig. 2 is the video steganalysis method flow chart provided by the invention based on intra prediction mode calibration.
Fig. 3 is the flow chart of intra prediction mode calibration of the invention.
The positional diagram of current block and adjacent block when Fig. 4 is compressed encoding.
Fig. 5 is fisrt feature collection distribution map of the invention.
Fig. 6 is second feature collection distribution map of the invention.
Specific embodiment
A specific embodiment of the invention is further described below in conjunction with attached drawing 3-6.
Video steganalysis method provided in an embodiment of the present invention based on intra prediction mode calibration includes intra prediction
Three model calibration, IPMC feature calculation and support vector machines study and classification big steps, concrete operations process are as follows:
Step 1: carrying out intra prediction mode calibration with 4 × 4 pieces for unit.It screens first and is possibly used for the 4 of insertion in I frame
× 4 pieces, then by these 4 × 4 pieces of block-by-block decompressions and second-compressed, 9 kinds of prediction modes are traversed during second-compressed and are counted
Calculate corresponding absolute transformed error and (SATD:Sum of Absolute Transformed Differences) (Kim, J.,
Jeong,J.:Fast Intra Mode Decision Algorithm using the Sum of Absolute
Transformed Differences.In:Proceedings of 2011 International Conference on
Digital Image Computing:Techniques and Applications,DICTA 2011,pp.655-
659.IEEE (2011)), SATD is defined as follows:
Wherein, A indicates that present encoding block, S presentation code block compress preceding pixel value set, SRECAfter indicating that decompression is rebuild
Sets of pixel values, (x, y) are pixel coordinates, and H (*) is Ha Deman transforming function transformation function.The implementation of step 1 is specifically subdivided into following four
Sub-step (refers to Fig. 3):
1) calibration will be minimum unit progress with 4 × 4 pieces in I frame, I frame be taken out from input video code stream, and take out I
One in frame be likely to be used for steganography insertion 4 × 4 pieces.It is empty in order to further save storage in H.264 compressed encoding
Between, positional relationship as shown in Figure 4, if the prediction mode of 4 × 4 pieces of C of present encoding is equal to block A adjacent and above and left block B prediction
When the smaller value of mode, encoder meeting setting flag position Pre=1 no longer needs to the prediction mode of storage current block C at this time.Therefore,
All steganographic algorithms based on intra prediction mode will not all select 4 × 4 pieces of Pre=1 to be embedded in.Therefore it is carried out in the present invention
Also 4 × 4 pieces of all Pre=1 will be skipped when calibration, remaining 4 × 4 pieces are thought to be likely to be used for steganography.
2) prediction mode and quantization parameter (QP:Quantization original in current 4 × 4 pieces are stored
Parameter), current 4 × 4 pieces of solutions are then depressed into airspace, obtain sets of pixel values.The step is according to H.264 encoder mark
Quasi- process carries out, and details repeats no more.
3) current 4 × 4 pieces of second-compressed, traverse 9 kinds of prediction modes and calculate corresponding SATD, then in compression process
Current 4 × 4 pieces of IPM-SATD calibration set is obtained, required reference pixel value can be from the adjacent block decompressed before
It is obtained in data, quantization parameter QP directlys adopt the original QP value stored in step 2).
4) sub-step 1 is repeated) to 3) until no available 4 × 4 pieces.
Step 2: IPMC feature set being extracted according to the related data recorded in step 1, IPMC feature set includes that IPM transfer is general
Two subsets of rate feature set and SATD transfer distance feature set, calculating process includes following two sub-step:
1) the intra prediction mode set calculating IPM transition probability according to each 4 × 4 pieces of calibration front and backs for participating in calibration is special
It collects (fisrt feature collection), characteristic dimension is 9.The IPM that this feature set features after calibration deviates the general of original optimal IPM
Rate, calculation formula are as follows:
Wherein, x ∈ [1,9] is the feature serial number of fisrt feature collection, and k is current frame number, LkBe participate in calibration 4 ×
4 pieces of sums.Il∈ [1,9] is original IPM value,It is x-th in current 4 × 4 pieces of IPM-SATD calibration set
IPM, in formula (2):
Fig. 5 comparison shows under different code rates the fisrt feature collection point of (0.2Mb/s and 1Mb/s) steganography and non-steganography video
Cloth situation, a kind of video steganographic algorithm based on intra prediction mode that we are proposed using scholar Bouchama in experiment
(Bouchama,S.,Hamami,L.,Aliane,H.:H.264/AVC Data hiding based on intra
prediction modes for real-time application.In:Proceedings of the World
Congress on Engineering and Computer Science, Vol.1, pp.655-658 (2012)) production steganography
Video sample.As can be seen that most probability is fallen in first feature for non-steganography video, this shows absolutely mostly
Several IPM is consistent with initial IP M after calibration.However for steganography video, hence it is evident that more probability have been distributed to other
In feature, that is to say, that the IPM after calibration has higher probability to deviate initial IP M.
2) SATD transfer distance spy is calculated according to each absolute error sum aggregate for participating in 4 × 4 pieces calibrated calibration front and backs is total
It collects (second feature collection), characteristic dimension is 4.This feature features the distance that the SATD after calibration deviates original value.SATD's
Deviation may be as caused by compression artefacts or IPM steganography, but SATD caused by compression artefacts deviate very little, especially code rate
Almost close to 0 when sufficiently high, and IPM steganography causes SATD deviation then to want relatively high more.The calculation formula of second feature collection is such as
Under:
Wherein, y ∈ [Isosorbide-5-Nitrae] is the feature serial number of second feature collection, DlIndicate original SATD,It is optimal after indicating calibration
The corresponding SATD value of IPM, in formula (4):
Second feature collection actually depicts discrete probability distribution of the SATD deviation distance on four sections.Using discrete
Probability distribution can effectively reduce characteristic dimension.In above formula, β is interval division parameter.In general, β and video
Code rate it is negatively correlated, table 1 gives the reference value of β under different code rates.
The reference value of β under the different code rates of table 1.
Fig. 6 comparison shows under different code rates the second feature collection point of (0.2Mb/s and 1Mb/s) steganography and non-steganography video
Cloth situation.Similar to fisrt feature collection, for non-steganography video, most probability are fallen in first feature, this shows
SATD deviation distance is generally less than normal.And for steganography video, SATD deviation distance obviously increases, and more probability have fallen in it
In its feature.
The above sub-step 1) and 2) in two feature sets being calculated merge up to final IPMC feature set, feature
Lump dimension is 9+4=13 dimension.
Step 3: being learnt and classified using SVM support vector machines, kernel function uses gaussian kernel function.If input is instruction
Practice Sample video, is then used for the IPMC feature of extraction to train classifier template.It is mentioned with frame group (generally setting one group of 8 frame) as unit
Feature is taken, it should be identical or as close as possible with the code rate of group training sample；If input is video to be checked, first according to be checked
Video code rate matches corresponding classifier template, then is used to classify to determine whether video is steganography by the IPMC feature of extraction
Carrier.Video code rate is the principal element for influencing detection performance, therefore either learns and classify all make video code rate
For important referential data.SVM support vector machines is the prior art, and it will not go into details for this specification.
Specific embodiment described herein is only to give an example to general principles of the present invention.Skill belonging to the present invention
The technical staff in art field can do the similar mode of various modify or supplement or adopt to described specific embodiment and replace
Generation, but without departing from essential concept of the invention or beyond the scope of the appended claims.
Claims (6)
1. a kind of video steganalysis method based on intra prediction mode calibration, step include:
1) video to be checked is subjected to intra prediction mode calibration, the intra prediction mode calibration is with 4 × 4 pieces for unit block-by-block solution
Second-compressed is pressed and carried out, various intra prediction mode IPM are traversed during second-compressed and calculates corresponding absolute transformed
Error and SATD；
2) intra prediction mode alignment features collection, the intra prediction mode calibration are calculated according to the data recorded in calibration process
Feature set includes two subsets of IPM transition probability feature set and SATD transfer distance feature set, the IPM transition probability feature
Collection deviates the probability of original optimal IPM for the IPM after calibration, and the SATD transfer distance feature set is that the SATD after calibration deviates
The distance of original value；The IPM is calculated according to the intra prediction mode set of each 4 × 4 pieces of calibration front and backs for participating in calibration to turn
Probability characteristics collection is moved, calculation formula is as follows:
Wherein, x ∈ [1,9] is the feature serial number of fisrt feature collection, and k is current frame number, LkBe participate in 4 × 4 pieces of calibration it is total
Number, Il∈ [1,9] is original IPM value,It is x-th of IPM in current 4 × 4 pieces of IPM-SATD calibration set, on
It states in formula:
3) the intra prediction mode alignment features collection is supplied to classifier to be learnt and classified, and then carries out steganography point
Analysis.
2. the method as described in claim 1, which is characterized in that step 1) includes following sub-step:
I frame 1-1) is taken out from input video code stream, and takes out one in I frame 4 × 4 pieces for being likely to be used for steganography insertion；
Prediction mode and quantization parameter original in current 4 × 4 pieces 1-2) are stored, current 4 × 4 pieces of solutions are then depressed into airspace,
Obtain sets of pixel values；
1-3) current 4 × 4 pieces of second-compressed, traverse 9 kinds of prediction modes and calculate corresponding SATD in compression process, then
It calibrates and gathers to current 4 × 4 pieces of IPM-SATD, required reference pixel value is obtained from the adjacent block data decompressed,
Quantization parameter directlys adopt step 1-2) in store original quantisation coefficient value；
1-4) repeat sub-step 1-1) to 1-3) until no available 4 × 4 pieces.
3. the method as described in claim 1, it is characterised in that: before and after step 2) is according to each 4 × 4 pieces of calibrations for participating in calibration
Absolute error sum aggregate it is total calculate SATD transfer distance feature set, calculation formula is as follows:
Wherein, y ∈ [Isosorbide-5-Nitrae] is the feature serial number of second feature collection, DlIndicate original SATD,Optimal IPM after indicating calibration
Corresponding SATD value, in above-mentioned formula:
In above formula, β is interval division parameter.
4. the method as described in claim 1, it is characterised in that: the step 3) classifier is SVM support vector machines, core letter
Number uses gaussian kernel function.
5. method as described in claim 1 or 4, it is characterised in that:, will if input is training sample video in step 3)
IPMC feature is for training classifier template；If input is video to be checked, corresponding classifier template is matched first, then will
IPMC feature is for classifying to determine whether video is steganography carrier.
6. method as claimed in claim 5, it is characterised in that: step 3) matches corresponding according to the video code rate of video to be checked
Classifier template.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201510438280.9A CN106375768B (en) | 2015-07-23 | 2015-07-23 | Video steganalysis method based on intra prediction mode calibration |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201510438280.9A CN106375768B (en) | 2015-07-23 | 2015-07-23 | Video steganalysis method based on intra prediction mode calibration |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN106375768A CN106375768A (en) | 2017-02-01 |
| CN106375768B true CN106375768B (en) | 2019-05-17 |
Family
ID=57881000
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201510438280.9A Expired - Fee Related CN106375768B (en) | 2015-07-23 | 2015-07-23 | Video steganalysis method based on intra prediction mode calibration |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN106375768B (en) |
Families Citing this family (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109711121B (en) * | 2018-12-27 | 2021-03-12 | 清华大学 | Text steganography method and device based on Markov model and Huffman coding |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101500161A (en) * | 2008-01-31 | 2009-08-05 | 华为技术有限公司 | Inter-frame prediction method and apparatus based on adaptive block transformation |
| CN103765892A (en) * | 2011-06-28 | 2014-04-30 | 三星电子株式会社 | Method and apparatus for coding video and method and apparatus for decoding video, accompanied with intra prediction |
| CN104125467A (en) * | 2014-08-01 | 2014-10-29 | 郑州师范学院 | Embedding and extracting methods for video steganography information |
| CN104602005A (en) * | 2010-08-17 | 2015-05-06 | M&K控股株式会社 | Method for decoding intra-predictions |
Family Cites Families (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US7822223B2 (en) * | 2006-01-13 | 2010-10-26 | New Jersey Institute Of Technology | Method for identifying marked images using statistical moments based at least in part on a JPEG array |
-
2015
- 2015-07-23 CN CN201510438280.9A patent/CN106375768B/en not_active Expired - Fee Related
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101500161A (en) * | 2008-01-31 | 2009-08-05 | 华为技术有限公司 | Inter-frame prediction method and apparatus based on adaptive block transformation |
| CN104602005A (en) * | 2010-08-17 | 2015-05-06 | M&K控股株式会社 | Method for decoding intra-predictions |
| CN103765892A (en) * | 2011-06-28 | 2014-04-30 | 三星电子株式会社 | Method and apparatus for coding video and method and apparatus for decoding video, accompanied with intra prediction |
| CN104125467A (en) * | 2014-08-01 | 2014-10-29 | 郑州师范学院 | Embedding and extracting methods for video steganography information |
Non-Patent Citations (1)
| Title |
|---|
| 基于转移概率矩阵的H.264/AVC视频帧内预测模式信息隐藏检测算法;孔维国，王宏霞，王科人，刘正辉;《四川大学学报》;20140630;全文 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN106375768A (en) | 2017-02-01 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN103345767B (en) | A kind of JPEG image steganography method of high security | |
| Lin et al. | An improved DCT-based perturbation scheme for high capacity data hiding in H. 264/AVC intra frames | |
| CN109635791B (en) | Video evidence obtaining method based on deep learning | |
| CN107197297B (en) | Video steganalysis method for detecting steganalysis based on DCT coefficient steganalysis | |
| CN106162195B (en) | A kind of 3D HEVC deep video information concealing methods based on single depth frame mode | |
| Nie et al. | Defining Embedding Distortion for Intra Prediction Mode-Based Video Steganography. | |
| CN102685500B (en) | H.264 large-capacity information hiding method based on information variable-length packet structure | |
| CN107657228A (en) | Video scene similarity analysis method and system, video coding-decoding method and system | |
| CN102857831B (en) | H.264 video integrality authentication method | |
| CN104780379A (en) | Compression method for screen image set | |
| CN103402087A (en) | Video encoding and decoding method based on gradable bit streams | |
| CN109982071B (en) | HEVC (high efficiency video coding) dual-compression video detection method based on space-time complexity measurement and local prediction residual distribution | |
| CN110312138B (en) | High-embedding-capacity video steganography method and system based on time sequence residual convolution modeling | |
| CN103826125A (en) | Concentrated analysis method of compressed surveillance video and device | |
| CN107547902B (en) | Adaptive rate distortion optimization method for surveillance video coding | |
| CN109819260A (en) | Video steganography method and device based on the fusion of multi-embedding domain | |
| CN105681803B (en) | A kind of HEVC video information hiding methods of large capacity | |
| CN103020138A (en) | Method and device for video retrieval | |
| CN111741313B (en) | 3D-HEVC rapid CU segmentation method based on image entropy K-means clustering | |
| CN106375768B (en) | Video steganalysis method based on intra prediction mode calibration | |
| CN106331730A (en) | Double-compression detection method by using quantification factor same as H.264 video | |
| Ouyang et al. | The comparison and analysis of extracting video key frame | |
| CN107454408A (en) | A kind of method of Image Coding code check dynamic adjustment | |
| CN111723735A (en) | Pseudo high bit rate HEVC video detection method based on convolutional neural network | |
| CN107682703B (en) | Video steganalysis method, device, equipment and computer readable storage medium based on the detection of inter-frame forecast mode recovery characteristic |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant | ||
| CF01 | Termination of patent right due to non-payment of annual fee | ||
| CF01 | Termination of patent right due to non-payment of annual fee |
Granted publication date: 20190517 Termination date: 20190723 |